# Knowledge Document Composer

> Turn a current provenance-checked source_analysis_pack or compatible video_analysis_pack into a gated report, outline, briefing, script, or research note with Source / Inference / Extension separation, composer/final receipts, and final quality checks.

- Skill: `sitabanubanu/knowledge-document-composer` (Agent Skill, multi-file: 10 files)
- Install (CLI): `npx skillmds@latest add sitabanubanu/knowledge-document-composer`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sitabanubanu/knowledge-document-composer/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: sitabanubanu (https://skillmd.com/u/sitabanubanu)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/sitabanubanu/knowledge-document-composer

---


# Knowledge Document Composer

Use this skill to turn structured source material into a finished document.

Core workflow:
1. Before processing any pack or derived artifact, require a current
   `gate_receipt.json` and `analysis_receipt.json`. File existence and an old
   `source_status` are not sufficient.
2. Read references/source-gate.md and apply it as a hard gate before writing commitments, source reconstruction, claim inventory, argument flow, or final prose.
3. Load only the source artifacts allowed by the source gate, such as `source_analysis_pack.md`, legacy `video_analysis_pack.md`, clean_transcript, concepts, examples, claims, source_logic, and logic_graph.
4. Read references/artifact-schema.md before writing intermediate or final artifacts.
5. Read references/workflow.md for the detailed stage process.
6. Prefer `kw audit --project-root <project>` so planning output is
   followed by `composer_receipt.json`. Direct runner output is not deliverable
   until that receipt binds it to the current analysis receipt.
7. Build a commitment file that captures the source status, source's core question, thesis, narrative spine, document goal, final language, audience, must-preserve evidence, and expansion boundaries.
8. Reconstruct the source's argument flow, examples, concept transitions, and language logic before drafting only when the source gate permits that reconstruction.
9. Build a claim map that separates Source, Inference, and Extension content, and never upgrade degraded or secondary material into Source claims.
10. Read references/report-template.md before drafting a report.
11. Prefer `kw compose --project-root <project>`. It validates the
   composer receipt, runs the final writer, and writes
   `final_report_receipt.json` bound to the quality gate and report hash.
12. Use scripts/final_report_auditor.py independently when checking a manually edited report candidate. It must reject secondary-only/degraded intake, unregistered Source claims, missing Source / Inference / Extension separation, and missing partial-scope labels.
13. Draft, critique, revise, and then read references/quality-gates.md before final delivery. If the user did not specify a final language, use the current conversation language and treat language mismatch as a quality failure.
14. Output the final document and supporting maps only in the report mode allowed by the source gate.

Runner guidance:
- Use scripts/document_composer_runner.py as the first document-composer runner for audited source workflows. It consumes `source_analysis_pack.md` when present, otherwise legacy `10_video/video_analysis_pack.md`, plus `00_source/source_status.json`, `05_gap_check/evidence_audit.json`, `05_gap_check/evidence_map.json`, `05_gap_check/claim_source_audit.json`, inventory, source logic, and gap check artifacts, then writes `20_document/composer_intake.json`, `commitments.md`, `source_reconstruction.md`, `claim_map.json`, `expansion_plan.md`, `report_outline.md`, and `quality_check.md`.
- Treat `document_composer_runner.py` as an intake-to-outline gate, not as a full writing engine. It intentionally does not create `draft_report.md` or `final_report.md`; those require the draft, critique, revision, and final quality-gate workflow.
- Use scripts/final_report_writer.py as the normal planning-to-final runner. It consumes `20_document/composer_intake.json`, `commitments.md`, `source_reconstruction.md`, `claim_map.json`, `expansion_plan.md`, and `report_outline.md`, then writes `draft_report.md`, `critique.md`, `revised_report.md`, `quality_gate.json`, refreshed `quality_check.md`, and, only if approved, `final_report.md`.
- Use scripts/final_report_auditor.py for an independent final gate. It writes `quality_gate.json` with `approved_for_final_report`, `report_scope`, gate statuses, blocking gates, checked files, and source claim ids used.
- If the upstream source status is blocked, failed, secondary-only, degraded, missing primary material, the evidence audit has errors, required evidence sidecars are missing, or `claim_source_audit.json` has blocking claims, do not use the normal document runner to create full-report planning artifacts.
- If `source_status` is `source_partial`, the final report may be approved only as a visible partial-scope report.
- If `source_status` is `secondary_only`, `source_blocked`, `source_failed`, or `degraded_report_only`, the final writer and auditor must not create a normal `final_report.md`.
- Do not read provider or `00_acquisition` output directly. Composer only consumes source-gated packs and planning artifacts.
- Do not repair source status or upgrade metadata/search/snippet artifacts into Source claims.
- Do not export or present a final report when its final receipt is missing,
  stale, or hash-mismatched, even if `final_report.md` exists.

Required reference order:
1. references/source-gate.md
2. references/artifact-schema.md
3. references/workflow.md
4. references/report-template.md
5. references/quality-gates.md